Structure-oriented edge-preserving smoothing in the frequency domain: Application to enhance 3D seismic data volumes

Abstract

We present a structure-oriented and edge-preserving algorithm for denoising seismic exploration poststack data volumes in the frequency domain. After transforming the 3D data into frequency-space domain we apply, for each frequency slice, a 1D edge-preserving smoothing filter which is guided spatially by the so-called gradient structure tensor. The algorithm can be efficiently implemented using linear convolutions to compute the gradient structure tensor and simple operations for the 1D edge-preserving filters. Examples using 3D seismic field data volumes show that random as well as coherent footprint noise can be significantly reduced with no loss of important structural information.